Voice/unvoiced/mixed excitation classification of speech using the autocorrelation of the output of an ADPCM system

A technique is proposed to distinguish between voiced, unvoiced, and mixed excitation (V/UV/M) speech frames. The technique uses the sampled autocorrelation on the bit stream of the CCITT adaptive differential pulse code modulation which is computed for input speech. The location of the first side lobe peak of the autocorrelation function can successfully discriminate between the V/UV/M speech frames. When the energy is used together with the autocorrelation, the correct identification is improved.<>